Exploring Accessory Dwelling Unit Development in San Francisco

General Data Exploration

Number of ADU Permits Issued Per Year

This simple bar chart shows that after a boom in ADU development between 2013 and 2020, the number of permits per year have steadily declined.

Note: The 2023 data only reflects permits through November.

Text(0.5, 1.0, 'ADU Permits by Year')

Number of ADU Units Permitted Per Year

It’s important to note that a single permit could be issued for multiple ADU units. This chart shows the number of actual units approved per year.

While a bit more useful than the previous chart, it largely tells the same story.

Text(0.5, 1.0, 'ADU Units by Year')

Number of Permits by Number of Proposed ADU Units

How common is it for one permit to propose multiple ADU units? This chart shows that while it is most common for ADU permits to propose one unit, many propose multiple units, all the way up to 16 units.

This indicates that not all ADU permits are for the stereotypical “granny flat” in the back yard. Developers are also using the 2013 ADU law to convert apartment building garage space into housing units.

Text(0.5, 1.0, 'ADU Permits by Number of Proposed Units')

Single vs. Multi-Unit Permits

In fact, this pie chart shows that nearly 2/3 of all ADU units are built with a multi-unit permit.

ADU Permits by Neighborhood

To begin understanding the spatial distribution of ADU development, this chart shows how many units have been developed in each neighborhood.

However, if you aren’t familiar with the city’s neighborhoods, this isn’t terribly useful.

Text(0.5, 1.0, 'ADU Units per Neighborhood')

Geospatial Data Exploration

Number of Units Per Permit

This map shows the location of ADU permits, colored by the number of units proposed. The lightest yellow indicates a single proposed unit, while the darkest red indicated four or more proposed units.

We can see that multi-unit ADU development is clustered in the city’s northeast quadrant, close to downtown. The western and southern halves of the city have a higher proportion of single-unit permits, indicating that these less dense neighborhoods have more traditional “granny flats.”

Make this Notebook Trusted to load map: File -> Trust Notebook

Number of Permits per Year

In this map, the darker the dot, the more recent the permit is. We can see that the same northeast quadrant shows a high proportion of relatively early ADU development, while the western and southern halves of the city show more recent development.

Make this Notebook Trusted to load map: File -> Trust Notebook

Taken together, these maps show that in the years immediately following the adoption of the ADU law, developers near downtown used the law as an opportunity to convert multiple apartment building garages into units. In recent years however, more traditional ADUs - single units in less dense neighborhoods - have become more common.

Demographic Determinants of ADU Development

In an effort to better understand what factors lead to ADU development in San Francisco, I pulled a number of Census variables to see whether any of them have a clear spatial relationship to ADU development.

The first map below shows total number of ADU Units in each of the city’s Census tracts.

The second map shows the various Census variables for each tract. Use the dropdown widget to explore different variables.

Finally, the scatter plot shows the relationship between each variable and the total number of ADU units, plotted with a linear regression line. Use the dropdown widget to explore different variables.

WARNING:param.main: column option not found for polygons plot with bokeh; similar options include: []

Ultimately, no clear geospatial trends emerge between the Census variables and ADU development. However, the scatterplots show us that there is a slight positive relationship between median rent and median income and ADU development, indicating that ADUs are more likely to be built in wealthier, more expensive neighborhoods. Conversely, there is a slight negative relationship between the total number of vehicles available in a Census tract and the total number of ADUs. This indicates that ADUs are more likely to be built in neighborhoods that are less car dependent.

Given that none of the included Census variables showed a strong, compelling relationship with ADU development, the next step in this work would be to widen the number of variables included and use those variables to conduct multivariate regression or machine learning to better understand and predict ADU development.